Where Playlists Come From
Spotify's playlist library spans moods, activities, genres, eras, and regional trends. These playlists are built in two fundamentally different ways. Editorial playlists like RapCaviar are crafted by human curators who specialize in particular genres or markets, giving them deep awareness of cultural moments and niche listening trends. Algorithmically generated playlists like Discover Weekly and Daily Mix rely on personalization models that analyze audio attributes and listening relationships to surface tracks a user is likely to enjoy.
Since 2017, Spotify has been combining both approaches into what the company calls "Algotorial" technology — playlists that start with editorial expertise and are then personalized at scale by machine learning.
The Editorial Seed
Algotorial playlists begin with a curator defining a specific user need. For example, a road trip playlist starts with the hypothesis that the user wants "familiar songs you know all the words to, and would sing along to."
That kind of judgment is difficult to encode algorithmically. A singable song might be a track that was on heavy rotation last summer, one with a catchy repeated refrain, or something recently featured in a TV show that triggers nostalgia. It's a quality listeners recognize instantly but struggle to define — exactly why human intuition is required at this step.
The editor gathers candidate tracks into what Spotify calls a "pool." This pool combines the curator's musical expertise with advanced search filtering and performance metrics showing which tracks have historically done well in the playlist. Because the editor isn't sequencing the final list in advance, the pool can include less obvious and less popular choices alongside crowd-pleasers, without needing to strike a universal balance for every listener.
Where Algorithms Take Over
Once the pool is assembled, the personalization engine selects the appropriate tracks and orders them for each individual listener. This becomes especially valuable for playlists spanning broad genres. The road trip pool might contain Pop, Indie, Rock, and Hip-Hop tracks, but each user receives a version optimized for their taste while remaining confident that every candidate track is "singable."
After personalization, the editor collaborates with Spotify's design team on the playlist's title, description, and imagery. The cover art can be personalized as well — a '60s Rock playlist might show a British Invasion artist to one user and a Surf Rock artist to another, displaying whichever performer the listener has the strongest affinity for.
The first product of these combined efforts was Songs to Sing in the Car, a personalized editorial playlist built entirely through the Algotorial pipeline.
Continuous Listener Feedback
Machine learning models analyze each user's listening history to predict which tracks they'll want next, then sequence those selections so the flow of the listening session feels natural.
The system also learns from every interaction with the playlist. Listens, skips, and saves to the library all feed into the recommendation engine, refining not only how tracks from the pool are used but also the listener's overall taste profile. This provides ongoing improvements both for individual users and for the recommendation system as a whole.
Scaling Curatorial Expertise
Spotify reports that more than 81% of listeners cite personalization as a favorite aspect of the service. Algotorial playlists represent an effort to scale the judgment of the editorial team so every listener receives a version of a curated experience tailored to their preferences, rather than a single one-size-fits-all list.
The hybrid model pairs the curator's cultural awareness with the algorithm's ability to process each listener's habits, creating playlists that are consistently relevant across a broad audience. Listeners can find examples of Algotorial-powered playlists throughout the Spotify app:
